Cambridge Analytica’s ‘mindfuck tool’ could be totally useless

langaweb_ cambridge analytica Langa Web Siti Internet Dinamici

As the Cambridge Analytica controversy lurches from revelation to revenge this week, there’s an underlying assumption that the company’s illicit sampling of Facebook data actually gave them an edge in influencing the outcome of the US election – and the Brexit vote. The experts we spoke to aren’t so sure. It’s possible that Cambridge Analytica actually made both campaigns’ ad targeting worse.

That’s certainly the view of social scientist Sandra Matz, who was part of the Cambridge Psychometric Institute team who researched links between personality tests and Facebook likes and inspired Aleksandr Kogan to develop the Cambridge Analytica research. “Targeting based on this research might have some impact on people who have no idea who they’re voting for but frankly, you’d probably get similar results with different targeting”.

Kosinski and Stillwell developed an app in 2012 – MyPersonality – which tested for the Big Five personality traits used in psychometric tests: openness, conscientiousness, extraversion, agreeableness and neuroticism. The app asked users to rate how strongly they agree or disagree with some statements. Then, users could opt-in to share their Facebook profile data which allowed the team to measure correlations between the Five values and their Facebook likes and posts (just like a little Cambridge Analytica).

The results were hardly surprising. Refining their models, the team showed that just 68 Facebook “likes” by a user could predict their skin colour, sexual orientation and political leanings – voting Republican or Democrat – with an accuracy averaging 90 percent.

Part of the problem involves the science-lite use of psychographics – a technique devised by the marketing industry from psychometrics. Psychometrics is a long-standing psychology tool that measures traits and personality.

“Psychometrics would be a powerful tool in marketing or elections if you could use them to segment a population into groups based on the Big Five – but you can’t,” explains psychologist Alan Redman. “So, you have to use something else that’s convenient – like the way Cambridge Analytica scraped Facebook data.”

Psychographics are based on the Big Five but are used to map consumers activities (like Cambridge Analytica did), interests and opinions (known as AIO variables) using shorter versions of psychometric tests, explains Mark Ritson, professor at Melbourne Business School. “To be honest, psychographics have been out of fashion in marketing for the past 20 years,” he explains. “Elections, however, are a phenomenally specific micro moment. You know exactly when the electorate makes its decision. If you could influence maybe five per cent of the electorate then geo-target to constituencies, you might have an effect.”

When a company’s HR department uses psychometrics to employ someone, the questionnaire typically involves anything between 100 and 150 questions while the average app uses 10-20. “So, the groups you get are quite crude and you could just as well use star signs,” Redman argues.
For some, Cambridge Analytica ’s use of quick questions in a personality app on top of Facebook’s own in-house targeting may actually have degraded the effectiveness of such advertising. “You measure psychometric traits through questionnaires. Then you try to predict questionnaire results with another data set – people liking brands or pages on Facebook. This seems like a reduction from Facebook ‘likes’ to five traits and that may have thrown away a lot of information. There is some trace of people’s personality present in social media data – but most of the time the data that gives advertisers a real boost is the information they get from cookies or from previous customer visits.”
Could the results be used to persuade, like it seems Cambridge Analytica did? Eckles conducted research at Facebook in 2010 and in 2012 to see if users could be persuaded to vote in forthcoming elections. Facebook placed banner ads at the top of certain users news feeds with messages including their friends intention to vote. Around 0.24 per cent of the people who saw the banner clicked through and the results suggest that the experiment directly increased turnout by about 90,000 people in 2012. “What Cambridge Analytica has done it’s not significantly more effective.”